Road Edge Detection And Extraction By Remote Sensing Images

As an essential part of geographic information,road information plays an important role in human’s daily life because it not only provides locative references for navigational objects, but also facilitates the urban traffic planning. Traditional road information extraction mainly relies on manual operation where massive resources are in need while the precision of extracting result can hardly meet the demand of traffic planning in nowadays. Recent years, with the development of the remote sensing technology, the remote sensing image, which have a lot merits such as high precision, real time update, and etc., gives a new method for road extraction. Therefore, extracting road information by remote sensing image is a promising theme for land administration and traffic planning.The purpose of this research is to explore the road extracting approach by remote sensing image.First of all, the remote sensing images are preprocessed for dividing the road and non-road area. Specifically, the preprocess increases the contrast between road and non-road area by image enhancement and decreases the noise by Gaussian smoothing filter. Also, the preprocess uses morphological approach to eliminate the negative effect of the vehicles, traffic marks and lane marks in the image.Secondly, in order to raise the precision of road extraction, this research proposes an 5*5 template as an improvement of Sobel operator. The optimal settings of each directional template are derived based on the Pascal’s triangle theory. The experiment result turns out that, the new method not only solved the directional limitation of the traditional Sobel operator, but also improved the denoising ability as well. In comparison with other operators, the proposed operator has better integrity and continuity in road edge extraction results, especially in road curve detection.Last, to deal with the situation where non-road edges still remain after the extracting process, this research analyzes the typical geometry features of road and non-road area. And select the space and the length-width ratio of the minimum exterior rectangle for a pixel block as two parameters; the thresholds are set for picking out the areas that do not have road geometry feature.